Cordic algorithm for FFT butterfly structure

[sougatavlsi]

Sir,I have written the code for FFT butterfly structure in Xilinx 14.1 and platform is VHDL.while writing the code i have faced the problem in handling complex functions as well as floating point numbers.After referring through some pdf's i came to know that i have to implement cordic algorithm in order to solve the prob.So now my doubt is that should i need to implement cordic algo first and then implement FFT bcoz i hv already written the code for FFT its check syntax is ok but while simulating may be due to tht complex functions it is nt showing proper output.....................what should i do now please help me?

 

[achaleus]

hello sougatavlsi,
you said you wrote fft, my question is how you generated twiddle factors.. if your N is fixed you no need to go for cordic algorithm.. pre calculate twiddle factors and store it in ram block, if input samples (N) is vector then you have to go through cordic algorithm

 

[sougatavlsi]

hi achaleus...i have generated twiddle factor on the basis of calculative formula....this is some portion of code which helps u to clarify the doubt.


architecture behav of fft_eightpt is
component butterfly is
port(x1,x2:in complex; --input
w: in complex; --twiddle factor
g1,g2ut complex); --output but acting as input to 2nd and 3rd stage
end component;

signal g1,g2 : comp_array := (others => (0.0,0.0));
constant w:comp_array2:=((1.0,0.0),(0.707,-0.707),(0.0,-1.0),(-0.707,-0.707));

 

[TrickyDicky]

I assume you do not intend to synthesise this code, as complex and real types are not synthesisable.

 

[sougatavlsi]

I assume you do not intend to synthesise this code, as complex and real types are not synthesisable.

yes sir, i agree the code will not be synthesizable since i am dealing with complex calculations...but my doubt is that is it really necessary to go for cordic in order to resolve the complex and flating point issues with FFT

 

[achaleus]

since your N is fixed you no need to go for cordic, represent twiddle factors in binary form.. and do calculation( represent real numbers(both input and twiddle factors) in some 16.16(Q.Q) format for example

0.707 rep in 0000000000000000.1011010011111101(remove decimal point) do some shifting operations according to the output while butterfly calculation)

 

[sougatavlsi]

so should i implement cordic algorithm first

 

[achaleus]

I said no need as your N is fixed just generate twiddle factors using some matlab or c.. save those values in a read only ram block